In Vitro Antibody Evolution Targeting Germline Hot Spots to Increase Activity of an Anti-CD22 Immunotoxin*

Recombinant immunotoxin BL22, containing the Fv portion of an anti-CD22 antibody, produced complete remissions in most patients with drug-resistant hairy cell leukemia but had less activity in leukemias with low CD22 expression. Complementarity-determining region (CDR) mutagenesis is used to increase antibody affinity but can be difficult to perform successfully. We previously showed that antibodies with increased affinity and immunotoxins with increased activity could be obtained by directing mutations at specific DNA residues called hot spots. Because hot spots can arise either by somatic mutation or be present in the germline, we examined which type of hot spot is preferred for increasing antibody affinity. Initially, a second generation antibody phage-display library targeting a germline hot spot (Ser 30 –Asn 31 ) within CDR1 of the antibody light chain was mutated. Substitution of serine 30 or aspara-gine 31 with arginine produced mutant immunotoxins with an affinity (0.8 n M ) increased 7-fold over BL22 (5.8 n M ) and 3-fold over the first generation mutant HA22 (2.3 n M ). More importantly, a 10-fold increase in activity over BL22 and a 2–3-fold increase over HA22 were observed in various B lymphoma cell lines including WSU-CLL that contains only 5500 CD22 sites per cell. For compar-ison, two phage-display libraries targeting non-germ-line hot spots in heavy chain CDR1 and CDR3 were generated but did not produce Fv with increased affinity. Our results demonstrate that germline hot spots but not non-germline hot spots are effective for in vitro antibody affinity maturation. 55 °C for 1 min, and 72 °C for 1 min. This reaction generated a 344-bp product that contained the mutations. The product was purified using a Qiagen Quick Spin column (Qiagen, Valencia, CA) and then quantitated by visualization on a 1% agarose gel. The purified products generated in the first PCR were used as primers in a second PCR. In this reaction, 2 pmol of the product from the first reaction was used with 20 pmol of the DNA oligomer S1 with 50 pg of template phagemid. The primers and template were mixed with two PCR beads in a 50- (cid:1) l volume and cycled using the profile described above. The reaction generated an 884-bp insert library. The PCR product was digested with SfiI and NotI and purified by using a QIAquick column (Qiagen). Purified PCR product (150 ng) was ligated with 250 ng of pre-cut phage-display vector pCANTAB 5E (Amersham Biosciences) and desalted. Forty ng of the ligation was used to transform TG1 elec-troporation-competent Escherichia coli cells (Stratagene). Each trans- formation produced a phage-display library containing (cid:3) 2 (cid:4) 10 4 independent clones. For LibVH96/97/98, five independent transforma- tions were performed to produce more than 1 (cid:4) 10 5 independent clones. The phage library rescued from the transformed bacteria, titered, and stored at 4 °C. identification Modeling of V Regions— The structural models of the variable region of RFB4 and its germline counterpart were generated by molecular modeling using web-based antibody modeling software WAM which uses a modified form of the algorithm used in antibody modeling software AbM of Oxford Molecular CA). In this study, we used the dead end elimination algorithm for side chain building VFF for screening. Molecular models viewed and analyzed using VMD SwissPdb Immunotoxins were added to the plate, and cells were incubated at 37 °C for 20 h, and the cell viability was measured with WST-8. Each assay was done in triplicate. IC 50 (mean values expressed in ng/ml) is the toxin concentration that reduced cell viability by 50% compared with the cells that were not treated with the toxin. The results are represented as means (cid:7) S.D. of triplicate determinations, and assays were repeated two or three times. In each CD22-positive cell line, repeated measures, analysis of variance with Dunnett’s and Student-Newman-Keuls post-tests, revealed a significant increase in cytotoxicity (IC 50 ) of E6, B8, D8, and B5 as compared with BL22, HA22, and the H11 and E12 mutants ( in Daudi and WSU-CLL p (cid:8) 0.05 The H11 and H12 mutants did not show increase of cytotoxicity as compared with HA22 ( p (cid:5) 0.05). No statistically significant differences among the B8, D8, and B5 immunotoxins were observed Repeated measures analysis of variance analysis with post-hoc Dunnett’s and Student-Newman-Keuls tests revealed that the E6, B8, D8, and B5 mutants significantly increased their activity by (cid:3) 2.8-, 3-, 2.5-, and 3-fold as compared with HA22 ( p (cid:8) 0.01). When tested under similar conditions, the H11 and H12 mutants did not show any increase of cytotoxicity ( p (cid:5) 0.05). No statistically significant differences among the E6, B8, D8, and B5 immunotoxins were observed ( p (cid:5) 0.05).

Recombinant immunotoxin BL22, containing the Fv portion of an anti-CD22 antibody, produced complete remissions in most patients with drug-resistant hairy cell leukemia but had less activity in leukemias with low CD22 expression. Complementarity-determining region (CDR) mutagenesis is used to increase antibody affinity but can be difficult to perform successfully. We previously showed that antibodies with increased affinity and immunotoxins with increased activity could be obtained by directing mutations at specific DNA residues called hot spots. Because hot spots can arise either by somatic mutation or be present in the germline, we examined which type of hot spot is preferred for increasing antibody affinity. Initially, a second generation antibody phage-display library targeting a germline hot spot (Ser 30 -Asn 31 ) within CDR1 of the antibody light chain was mutated. Substitution of serine 30 or asparagine 31 with arginine produced mutant immunotoxins with an affinity (0.8 nM) increased 7-fold over BL22 (5.8 nM) and 3-fold over the first generation mutant HA22 (2.3 nM). More importantly, a 10-fold increase in activity over BL22 and a 2-3-fold increase over HA22 were observed in various B lymphoma cell lines including WSU-CLL that contains only 5500 CD22 sites per cell. For comparison, two phage-display libraries targeting non-germline hot spots in heavy chain CDR1 and CDR3 were generated but did not produce Fv with increased affinity. Our results demonstrate that germline hot spots but not non-germline hot spots are effective for in vitro antibody affinity maturation.
Hematological malignancies remain a major public health problem. In the past several years, immunotoxins have been developed as a new therapeutic approach to treat these malignancies. Immunotoxins are composed of antibodies or their Fv biochemically or genetically linked to toxins, toxin subunits, or ribosome-inactivating proteins from bacteria, plants, or fungi (reviewed in Refs. [1][2][3]. The antibody portion binds to the antigen expressed on the target cell, and the toxin is internalized and causes cell death by arresting protein synthesis and inducing apoptosis. Hematological malignancies are attractive targets for immunotoxin therapy because the tumor cells are easily accessible to the therapeutic agent (4). CD22 is a lineage-restricted B cell antigen expressed in about 70% of B cell lymphomas and leu-kemias. Because CD22 is not expressed on stem cells and in the early stages of B cell development (5), it is an attractive target for cancer therapy. Clinical trials have been conducted with RFB4(dsFv)-PE38 (BL22), a recombinant immunotoxin containing an anti-CD22 murine monoclonal antibody RFB4 (6) Fv fused to a truncated form of Pseudomonas exotoxin A (PE38). BL22 was engineered by inserting a disulfide bond in the framework region so that it holds the Fv together but does not interfere with binding to antigen (7,8). BL22 is able to kill leukemic cells from patients and induced complete remissions in mice bearing lymphoma xenografts (9,10). BL22 has been evaluated in a phase I clinical trial at the NCI, National Institutes of Health, in patients with hematological malignancies. Sixteen patients with purine analogue-resistant hairy cell leukemia (HCL) 1 were treated with BL22 and 11 (86%) achieved complete remissions (11). BL22 is the first agent that is able to induce a high complete remission rate in patients with purine analogue-resistant HCL and confirms the concept that immunotoxins can produce clinical benefit to patients with advanced malignancies. Because of the clinical benefits obtained with BL22, we decided to improve this molecule by increasing its affinity and consequently its activity. This should lead to an increase in its activity in patients with malignancies such as chronic lymphocytic leukemia (CLL), in which the cells have relatively small amounts of CD22 on the cell surface (10,12).
Efforts have been made to improve the affinity of antibodies by random or semi-random mutation (13)(14)(15)(16)(17)(18)(19). Recently, a strategy using hypermutating B cell lines mimicking the natural somatic hypermutation has been developed (20). We have developed an approach mimicking somatic hypermutation to increase antibody affinity by using small antibody phage-display libraries (21). We used a few hot spots in the antibody complementarity determining regions (CDR) that are naturally prone to hypermutations (22). Hot spots RGYW (R ϭ A or G, Y ϭ C or T, and W ϭ T or A) are DNA sequences that are frequently mutated during the in vivo affinity maturation of an antibody (22)(23)(24)(25). We introduced random mutations around these sites by PCR and made phage-display libraries with a minimum size of only 10 3 and 10 4 independent clones (26). Panning of these small hot spot libraries has yielded mutant immunotoxins with an increase in activity (21,26,27). One of these is immunotoxin HA22 that has a mutation in a hot spot region of heavy chain CDR3 (27). We hypothesized that a directed evolution approach targeting a different hot spot in each generation would allow us to sequentially increase the activity of immunotoxins. This uphill-climbing strategy has the major advantage that in each generation the library size is small and easy to make, whereas * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  it is technically very difficult to produce large phage libraries where many nucleotides and amino acids in the Fv are randomly mutated. Because large changes in antibody affinity can result from accumulating point mutations, each with small effects (28), the capability to identify slight improvements in each generation could be of significant value. Here we have used immunotoxin HA22 as a starting point, and we examined the effect of mutating hot spots in germline residues or nongermline residues that have already undergone somatic mutation. We observed that mutation of germline hot spots but not non-germline hot spots resulted in Fv with increased affinity and immunotoxins with increased activity.

EXPERIMENTAL PROCEDURES
Cell Culture-Daudi, Namalwa, Ramos, MCF7, and 293T cells were obtained from American Type Culture Collection (Manassas, VA) and grown as described previously (8,27). The WSU-CLL line was kindly provided by Dr. A. Al-Katib (Wayne State University, Detroit, MI) and grown as described (29).
Construction of Phage-display Libraries-In the present study, we constructed 3-s generation phage-display libraries by mutagenizing hot spots in light chain CDR1 and heavy chain CDR1 or CDR3 of the HA22 Fv portion. HA22 Fv was a first generation mutant of RFB4 obtained by phage-display targeting the antibody heavy chain CDR3 hot spots (27). DNA oligomers were designed to generate a library randomizing 6 or 9 nucleotides (2 or 3 consecutive amino acids). Degenerate oligomers with the sequence NNS were used (N randomizing with all four nucleotides and S introducing only C or G).
The phagemid pCANTAB5E-G3 for display of HA22 single chain Fv (scFv) on the surface of bacteriophage M13 was made according to a previous procedure (27). The vector was modified by inserting a BspHI site (TCA TGA) containing a stop codon (underlined) at the hot spot positions using site-directed mutagenesis (QuickChange site-directed mutagenesis kit; Stratagene, La Jolla, CA). The resulting phagemids pCANTAB5E-G3-VL30/31BspHI, pCANTAB5E-G3-VH30/31BspHI, and pCANTAB5E-G3-VH96/97/98BspHI were used as a template to introduce two or three amino acid randomizations in a two-step PCR (26). The following oligonucleotides were used: S1, 5Ј-CAACGTGAAA-AAATTATTATTCGC-3Ј; RFB4VL30/31, 5Ј-AGTCAGGACATTNNSNN-STATTTAAACTGG-3Ј; RFB4VH30/31, 5Ј-GGATTCGCTTTCNNSNNS-TATGACATGTCT-3Ј; RFB4VH96/97/98, 5Ј-TACTGTGCAAGACAT-NNSNNSNNSGGTACGCACTGGGGG-3Ј; and AMBN, 5Ј-GCTAAA-CAACTTTCAACAGTCTATGCGGCAC-3Ј. In the first PCR, 50 pg of the phagemid pCANTAB5E-G3-VL30/31BspHI, pCANTAB5E-G3-VH-30/31BspHI, or pCANTAB5E-G3-VH96/97/98BspHI was used as the template in each library using 20 pmol of DNA oligomers AMBN along with 20 pmol of the DNA oligomers RFB4VL30/31, RFB4VH30/31, or RFB4VH96/97/98. The template and oligonucleotides were mixed with two Ready-To-Go PCR beads (Amersham Biosciences) in a 50-l volume and then cycled using the following profile: 1 cycle at 95°C for 5 min, followed by 30 cycles at 94°C for 1 min, 55°C for 1 min, and 72°C for 1 min. This reaction generated a 344-bp product that contained the mutations. The product was purified using a Qiagen Quick Spin column (Qiagen, Valencia, CA) and then quantitated by visualization on a 1% agarose gel. The purified products generated in the first PCR were used as primers in a second PCR. In this reaction, 2 pmol of the product from the first reaction was used with 20 pmol of the DNA oligomer S1 with 50 pg of template phagemid. The primers and template were mixed with two PCR beads in a 50-l volume and cycled using the profile described above. The reaction generated an 884-bp insert library. The PCR product was digested with SfiI and NotI and purified by using a QIAquick column (Qiagen). Purified PCR product (150 ng) was ligated with 250 ng of pre-cut phage-display vector pCANTAB 5E (Amersham Biosciences) and desalted. Forty ng of the ligation was used to transform TG1 electroporation-competent Escherichia coli cells (Stratagene). Each transformation produced a phage-display library containing ϳ2 ϫ 10 4 independent clones. For LibVH96/97/98, five independent transformations were performed to produce more than 1 ϫ 10 5 independent clones. The phage library was rescued from the transformed bacteria, titered, and stored at 4°C.
Biopanning on Cells-Biopanning was performed on MCF7 cells to remove phage that bind to cells nonspecifically and CD22-positive Daudi cells to enrich for CD22-binding phage. For phage absorption, MCF7 cells (Ͼ5 ϫ 10 7 ) were pelleted and resuspended in 10 ml of cold blocking buffer (Dulbecco's phosphate-buffered saline ϩ 5% BSA ϩ 0.1% NaN 3 ). Phage (ϳ1 ϫ 10 12 plaque-forming units) from each library were added to the cell suspension, and the mixture was rotated slowly at 4°C for 90 min. The cells were then pelleted, and the supernatants (10 ml) containing unbound phage were incubated with Daudi cells (1 ϫ 10 6 ) and rotated slowly for 90 min at 4°C. The cells were then pelleted and resuspended in 10 ml of cold blocking buffer. The cells were then washed six times with 1 ml of cold blocking buffer for the first and second round of panning. At the third panning, the cells were washed 16 times. Bound phage were eluted by resuspending the washed cells in 0.5 ml of ice-cold 50 mM HCl (pH 1.5) and incubating on ice for 10 min. Daudi cells were pelleted, and the eluted phages were transferred to a tube containing 100 l of 1 M Tris (pH 8). The eluted phages were titered to determine the number of phage captured. The eluted phage (ϳ0.6 ml) were then amplified by reinfecting E. coli TG1 for use in the next round of panning.
Preparation and Purification of CD22-The extracellular domain of CD22 protein was expressed as a fusion to human IgG Fc in transfected 293T cells containing pCDNA1.1-22-Fc, a plasmid encoding the extracellular domain of human CD22, as described previously (27). The plasmid was transfected into 293T cells by Lipofectamine reagent (Invitrogen), and the CD22-Fc protein was harvested from the culture supernatant and purified with Hi-Trap protein A column (Amersham Biosciences).
Enzyme-linked Immunosorbent Assay (ELISA)-Phages or scFv were prepared from single colonies of E. coli TG1 containing phagemids selected in panning. In order to normalize the scFv expression, we performed a dot blot analysis in which serial dilutions of phage or scFv were spotted on a membrane. The amount of scFv was detected with an anti-E TAG antibody (Amersham Biosciences). The colorimetric intensity of each dot was measured with the NIH Image program (rsb.info. nih.gov/nih-image/Default.html). ELISA were performed using CD22-Fc fusion proteins. Briefly, 96-well ELISA plates (Maxi-sorb; NUNC, Rochester, NY) were coated with goat anti-human IgG (Jackson Immu-noResearch, West Grove, PA) (0.5 g/well) in 50 mM bicarbonate buffer (pH 9.5) at 4°C overnight, washed, blocked with 1% of BSA in phosphate-buffered saline (PBS) at room temperature for 1 h, and washed with PBS containing 0.05% Tween 20 (PBST). CD22-Fc fusion proteins, 5 g/ml in PBST containing 1% BSA, were added at 50 l/precoated well. After 1 h at room temperature, the plates were washed four times with PBST. Each well was then incubated for 1 h with 50 l of a 1:10 dilution of culture supernatant containing soluble scFv or phage clones (ϳ10 9 phages/well). After washing four times with PBST, each well was incubated with horseradish peroxidase-conjugated anti-E tag (for scFv) or anti-M13 (for phages) antibody (Amersham Biosciences) at room temperature. Assays were developed with tetramethylbenzidine/H 2 O 2 substrate (Pierce). After 15 min of incubation in the dark, the reaction was stopped with 100 l of 1 N sulfic acid, and the absorbance was read spectrophotometrically at 450 nm.
Flow Cytometry-Phage were prepared from single colonies of E. coli TG1 containing phagemids selected after panning. Flow cytometry was carried out as follows: 5 ϫ 10 5 Daudi cells were incubated with 1 ϫ 10 12 phage or scFv (ϳ10 g/ml) at room temperature for 60 min; cells were washed two times with blocking buffer containing 5% BSA plus 0.1% NaN 3 in PBS, and 5 g of mouse anti-M13 antibody or anti E-Tag antibody (Amersham Biosciences) was added to each sample. The mixture was incubated at room temperature for 20 min and then washed twice with blocking buffer. A goat antimouse fluorescein isothiocyanate-labeled antibody (Jackson Immu-noResearch) was added, and cells were incubated for 20 min at room temperature. Cells were washed two times, and analysis was performed in a FACSCalibur machine (BD Biosciences). Data were acquired using Cell Quest software (BD Biosciences).
Construction and Expression of Immunotoxin-scFv from selected phagemids were PCR-amplified using primers that introduced NdeI and HindIII restrictions sites. The products of the reaction were purified, digested with NdeI and HindIII, and cloned into a T7 expression vector in which the scFv was fused to a truncated version of Pseudomonas exotoxin A (PE38). The expression and purification of recombinant immunotoxins was performed as described previously (26).
Cytotoxicity Assays-Cytotoxicity on cell lines was measured by both protein synthesis inhibition and cell death assays. In both assays, cells were plated in 96-well plates at a concentration of 5 ϫ 10 4 cells/100 l/well. Immunotoxins were serially diluted in PBS, 0.2% human serum albumin, and 20 l of the resulting mixture was added to each well. Plates were incubated for 20 h at 37°C.
Protein synthesis was measured by [ 3 H]leucine incorporation. The cells were pulsed with 1 Ci/well [ 3 H]leucine in 20 l of PBS, 0.2% human serum albumin for 2.5 h at 37°C. Radiolabeled material was captured on filter mats and counted in a Betaplate scintillation counter (Amersham Biosciences). Triplicate sample values were averaged, and the inhibition of protein synthesis was determined by calculating percentage of incorporation compared with control wells without added toxin. The activity of the molecule is defined by IC 50 , i.e. the toxin concentration that reduced incorporation of radioactivity by 50% compared with the cells that were not treated with the toxin. All experiments were performed in triplicate on three separate occasions.
Cell death was assessed by WST-8 conversion using the Cell Counting Kit-8 according to recommendations from the supplier (Dojindo Molecular Technologies, Gaithersburg, MD). A volume of 10 l of WST-8 (5 mM WST-8, 0.2 mM 1-methoxy-5-methylphenazinium methylsulfate, and 150 mM NaCl) was added to each well, and the incubation was carried out for 4 h at 37°C. The absorbance of the sample at 450 nm was measured with a reference wavelength of 650 nm. Cytotoxicity was expressed as 50% inhibition of cell viability, which is halfway between the level of viability in the absence of toxin and that in the presence of 10 g/ml of cycloheximide. All experiments were performed in triplicate on two or three separate occasions. Statistical analyses were performed with Prism (version 3.02) for Windows (GraphPad software, San Diego, CA). Within each cell line, raw data were analyzed by application of one factor (treatment) repeated measures analysis of variance with Dunnett's and Student-Newman-Keuls post-tests. The confidence interval was set at 95% for all statistical tests. p values less than 0.05 was considered statistically significant.
Determination of Affinity Constants (K D )-Affinity measurements were performed with flow cytometry following the protocols described previously (30,31) except that in this study Daudi cells and monomeric immunotoxins instead of the whole antibody or scFv molecules were used. Equilibrium constants and Scatchard plots (32) were determined by using the Marquardt-Levenberg algorithm for nonlinear regression with the Prism software (version 3.0.2, GraphPad software).
Sequencing Analysis-The scFv fragments present in selected phage clones were sequenced using the dideoxy method with primers S1 and AMBN. DNA sequencing was performed using PE Applied Biosystems Big Dye Terminator Cycle Sequencing kit. The samples were run and analyzed on a PE Applied Biosystem model 310 automated sequencer (NCI DNA Sequencing Core Facility, Bethesda).
Both the heavy and light chains variable region sequences were numbered following the Kabat rule (33). The scFv sequences were analyzed by online V-Quest software provided by the International ImMunoGeneTics data base (IMGT, imgt.cines.fr/textes/vquest/) for identification of germline origin of V H /V L regions (34).
Structural Modeling of V Regions-The structural models of the variable region of RFB4 and its germline counterpart were generated by molecular modeling using web-based antibody modeling software WAM (35) (antibody.bath.ac.uk/index.html), which uses a modified form of the algorithm used in antibody modeling software AbM of Oxford Molecular (Accelrys, San Diego, CA). In this study, we used the dead end elimination algorithm for side chain building and VFF screen for final screening. Molecular models were viewed and analyzed using VMD (www.ks.uiuc.edu/) and SwissPdb Viewer (www.expasy.org/spdbv/).

Sequence Analysis of the RFB4 Antibody and Identification of DNA Hot Spots for in Vitro Affinity
Maturation-High affinity antibodies were generated in mice and humans by somatic hypermutation. Somatic hypermutation does not occur randomly within immunoglobulin V genes but is preferentially targeted to certain nucleotide positions (hot spots) and away from others (cold spots) (23,36). Cold spots often coincide with residues essential for V gene folding. Hot spots, which appear to be strategically located to favor affinity maturation, are most frequently located in the CDRs (particularly CDR1). This process mainly results in the introduction of mutations that are located at or very near (A/G)G(C/T)(A/T) (RGYW) and AG(C/T) (AGY) sequences. A total of 50 -60% of all somatic hypermutations are found around RGYW motifs (24). To improve the affinity of the antibody portion of recombinant immunotoxins and thereby increase their cytotoxic activity, we have proposed that hot spots are the preferred regions to target new mutations (21). The first step in this process is to identify the DNA hot spot sequences in all the CDRs of the antibody under investigation.
A homology search using V-Quest at IMGT (imgt.cines.fr/ textes/vquest/) (34) shows that the light chain variable region of antibody RFB4 used to make immunotoxin BL22 is of the V10 class and has 96% identity at the amino acid level with the germline light chain IGKV10-96*01 (GenBank TM accession number M15520). The light chain variable region gene probably originated from recombination of the V gene (IGKV10-96*01) and the J1 gene IGKJ1*01 (GenBank TM accession number V00777) without a somatic insertion. The DNA sequence of the germline V and J gene fragments were manually assembled and translated into an amino acid sequence. Fig. 1 showed alignment of the germline sequence with the amino acid sequence of the RFB4 V L region.
The heavy chain variable region of RFB4 was found to be of highest homology with the germline V H gene IGHV5S21*01 (GenBank TM accession number AF120463). It has 94% amino acid identity with this germline sequence. The C-terminal region probably originated from recombination of the germline D gene IGHD-FL16.1*01 (GenBank TM accession number J00434) and J gene IGHJ3*01 (GenBank TM accession number V00770) with an insertion resulting in one additional amino acid at the 5Ј-end and three amino acids at the 3Ј-end of the D gene. The germline V, D, and J gene fragments were manually assembled, converted into the amino acid sequence, and aligned with the RFB4 V H region (Fig. 1).
We identified two types of hot spot residues in the nucleotide sequences of the CDR domains of RFB4. One type contains hot spots in germline residues; the other type contains hot spots in residues that have already undergone mutational events. Germline hot spot residues have the same nucleotide sequence as the germline gene, whereas in non-germline hot spot residues the nucleotide sequence at or very near the RGYW motifs differs from the germline gene. The non-germline hot spots were presumably produced during somatic hypermutation of B cells. Five germline hot spot clusters (Fig. 1, red) were identified in the CDR domains of RFB4. Among them, four contain RGYW in their nucleotide sequences. They are located in light chain CDR1 (L1) and CDR3 (L3), and heavy chain CDR2 (H2) and CDR3 (H3) (Fig. 1). Three non-germline hot spot clusters were also found. All of them contain RGYW in their nucleotide sequences. They are located in heavy chain CDR1, CDR2, and CDR3 ( Fig. 1, green).
We generated a three-dimensional molecular model (Fig. 2) for the RFB4 V region using the WAM program. Ribbon diagrams of the superimposed germline and RFB4 V regions ( Fig.  2A) demonstrates that heavy chain CDR3 (H3) of RFB4 shows significant displacement from that of the germline, suggesting that heavy chain CDR3 is directly involved in antigen binding. A surface view of the RFB4 V region reveals that most germline (Fig. 2, red) and non-germline (green) CDR hot spot residues are exposed, suggesting that they may be located at the interface between the antibody and antigen (Fig. 2B). Ser H100 , Ser 100a , and Tyr 100b are highly exposed on the tip of heavy chain CDR3 loop (H3).
Selection of a Light Chain Germline Hot Spot in CDR1-As shown in the primary sequence alignment (Fig. 1), almost all the somatic mutations are located within the heavy chain CDR regions of antibody RFB4 except for a change of germline residue Arg 53 to Ile 53 in the light chain of CDR2. In the heavy chain CDR1, germline Thr 28 , Ser 31 , and Ala 33 were mutated to Ala 28 , Ile 31 , and Asp 33 . In the heavy chain CDR2, Asp 55 -Tyr 56 -Ile 57 and Ala 60 were mutated to Gly 55 -Thr 56 -Thr 57 and Pro 60 . In the heavy chain CDR3, in addition to somatic insertion of five amino acids (Ser 96 , Gly 97 , Gly 100c , Val 100d , and Leu 100e ) located at the ends of the germline D gene fragment, Arg 95 was mutated to His 95 . In the present study, we first chose a light chain germline hot spot in CDR1 for in vitro mutagenesis because unlike heavy chain CDR regions and light chain CDR2, light chain CDR1 and CDR3 have not been changed by somatic mutations of B cells.
Construction of a Phage-display Library Targeting a Germline Hot Spot in Light Chain CDR1-In this work, we used the Fv portion of HA22, a first generation mutant of monoclonal antibody RFB4, in which the affinity of the Fv had already been increased by mutating Ser 100 -Ser 100A -Tyr 100b to Thr 100 -His 100A -Trp 100b in a hot spot of the heavy chain CDR3 (27). The light chain Fv of HA22 is the same as that of BL22. We then constructed a phage display library (RFB4LibVL30/31) that targeted a germline hot spot (Ser 30 -Asn 31 ) located in the light chain CDR1 (L-CDR1) of HA22.
The new library was transduced into E. coli TG1 cells by electroporation and the Fv expressed as a fusion with the minor FIG. 2. A, ribbon diagram of RFB4 variable regions (orange) is superimposed with that of its germline counterpart (blue). H1, -2, and -3, heavy chain CDR1, -2, and -3; L1, -2, and -3, light chain CDR1, -2, and -3. Heavy chain CDR3 for RFB4 shows significant displacement from that of the germline one. B, a space-filling model of RFB4 Fv. The view is looking down on the antigen-binding surface. CDR1, -2, and -3 for both chains of RFB4 are colored (orange, red, green, or blue). Somatic mutations are either orange or green. Among them, the non-germline hot spot residues created during the somatic process are green. Germline hot spot residues within CDR domains, which were not changed in somatic mutations, are red. All residues are numbered according to the Kabat numbering scheme (33).  Anti-CD22 Immunotoxins 611 phage coat protein pIII after rescue with a helper phage. The library complexity of RFB4LibVL30/31 consisted of more than 1 ϫ 10 4 independent clones. Because it is only necessary to produce a library of 1024 (32 ϫ 32) independent clones to randomly mutate the hot spot sequence (AGCAAT) to NNSNNS, the size of the current library is 10-fold larger than the one required. Sequence analysis of 10 independent clones from the mutant library demonstrated that the targeted hot spot residues were successfully randomized (data not shown). Subtractive Biopanning and Analysis-To mimic somatic hypermutation in the immune response, which optimizes the differential binding affinity for an antigen relative to other molecules, we decided to screen for high affinity phage that could bind to CD22-positive Daudi but not CD22-negative MCF7 cells. To reduce the number of phage that bound nonspecifically, subtractive panning was performed first on CD22negative MCF7 cells, and then enrichment was performed on CD22-positive Daudi cells as described in detail under "Experimental Procedures." At the last (third) round of panning, the Daudi cells were washed 16 times with the blocking buffer. Each wash included an incubation for Ͼ15 min on ice. The total duration of the off-rate selection was 6 h at the third round of biopanning. Then bound phage were eluted (pH 1.5) ( Table I, RFB4LibVL30/31). The overall enrichment for library RFB4-LibVL30/31 was very large (10,700-fold). Fluorescence-activated cell sorting (FACS) analysis of the bound phage was used to monitor the enrichment of high binders after each biopanning step (Fig. 3A). The gradual enrichment of phage suggests that a small number of high affinity binders existed in the primary library RFB4LibVL30/31 and were gradually enriched during the process of biopanning.

Anti-CD22 Immunotoxins
In addition, we also conducted a phage ELISA on CD22, CD25, and CD30-Fc fusion proteins (Fig. 4A). The ELISA signals on CD22-Fc were increased after each round of panning, but the nonspecific background signals on CD25-Fc or CD30-Fc proteins were not. The results indicate that phages with a high affinity for CD22 were successfully enriched during biopanning.
After the third round of biopanning, monoclonal scFvs were prepared from individual phage clones and tested for their ability to bind to recombinant single chain CD22-Fc by ELISA (Fig. 5). Fig. 5, 1st panel, shows the CD22 binding activity of clones selected after panning of RFB4LibVL30/31. Thirty three clones were tested and could be grouped into four types as follows: 1) 23 clones had high ELISA signals; 2) 4 clones had ELISA signals comparable with HA22 scFv phage (Fig. 5, arrow); 3) 4 clones had ELISA signals lower than the starting parental phage; and 4) 2 clones had no signal. Twenty seven ELISA high binders in groups 1 and 2 were further analyzed on Daudi cells by flow cytometry (data not shown). Sixteen mutant scFvs with FACS signals (mean fluorescence intensity) for CD22-positive cells similar to or higher than HA22 and with high phage ELISA signals (Fig. 5) were selected for sequence analysis. The results are shown in Table II. The most frequent residues were Gly 30 -Arg 31 and Arg 30 -Gly 31 . Each of these appeared six times. His 30 -Gly 31 , Ser 30 -Gly 31 , Arg 30 -Ser 31 , and Ala 30 -Arg 31 each occurred once. More than one codon was used in the clones containing Gly 30 -Arg 31 or Arg 30 -Gly 31 , indicating that there was a strong selection for these amino acid residues. Furthermore, basic amino acids such as Arg and His were conserved in all the Fv that bound strongly to CD22 except clone B6.
In addition to the binding affinity, we also examined the specificity of representative mutant phage antibodies by measuring their binding to MCF7 cells. When tested on CD22negative MCF7 cells, some mutants showed reduced nonspecific binding as compared with HA22 (Table III), indicating that subtractive biopanning probably enriches for mutants that have lower nonspecific binding than the starting molecule. Based on these results, we decided to choose four Fv clones, E6 (Gly-Arg), B8 (Arg-Gly), B5 (His-Gly), and D8 (Ala-Arg), to make immunotoxins and to examine their binding and biological properties.
Construction, Expression of the Immunotoxins-We subcloned DNA encoding the Fv into a T7 promoter-driven expression vector in which the scFv is fused to a truncated version of Pseudomonas exotoxin A (PE38) to generate a recombinant immunotoxin. These were expressed in E. coli BL21 (DE3) cells, and upon induction with isopropyl-1-thio-D-galactopyranoside, large amounts of recombinant protein accumulated as intracellular inclusion bodies. We then refolded immunotoxins from solubilized inclusion bodies in a redox-shuffling buffer and purified them by ion-exchange chromatography on Q-Sepharose and MonoQ columns followed by TSK gel filtration chromatography. Purified recombinant immunotoxin monomers with the expected size of 63 kDa were obtained as analyzed by SDS-PAGE under nonreducing conditions (Fig. 6). The purity of each recombinant immunotoxin is above 90%.
Cytotoxic Activity-The recombinant immunotoxins were tested on several CD22-positive cell lines. Cytotoxicity (IC 50 ) was evaluated by both cell death assays (WST-8) in three Burkitt lymphoma cell lines (Daudi, Namalwa, and Ramos) and one B-CLL cell line (WSU-CLL) (Table IV) (Fig. 7) and Raji, another Burkitt lymphoma cell line (data not shown). Immunotoxins containing the E6, B8, B5, and D8 Fv light chains were 2-or 3-fold more active than HA22 and up to 10-fold more active than BL22 in different B cell lines. Among these mutants, E6 and B8 Fv were found in 6 of 16 clones analyzed after biopanning, whereas D8 and B5 were found only once (Table II). Nevertheless, a common pattern of increase in their immunotoxin activities is present. More importantly, as shown in Table IV, in WSU-CLL, a cell line with low CD22 expression established from a B-CLL patient resistant to chemotherapy (29), the mutant immunotoxins (E6, B8, D8, and B5) were significantly more active than HA22 and BL22. Daudi cells have 1 ϫ 10 5 CD22 sites/cell (37), but WSU-CLL cells have only 5500 CD22 sites/cell (10). None of the mutant immunotoxins were cytotoxic to the CD22-negative cell line MCF-7, which demonstrates that the cytotoxic effect of these immunotoxins is selective to antigen-positive cells.

and protein synthesis inhibition ([ 3 H]leucine incorporation) in Daudi cells
Finally, we performed affinity measurement on Daudi cells (Table IV) by using monomeric recombinant immunotoxins. As predicted from the increase in cytotoxic activity, the binding affinity of mutants E6, B8, D8, and B5 for native CD22 on cells is increased from 2.3 to 0.8 nM. The increase in cytotoxicity was generally proportionally related to the increase in affinity.
Phage Display Libraries Targeting Non-germline Hot Spots-The results with phage library RFB4LibVL30/31 and previous data (21,27) indicate that germline hot spot residues are good candidates for in vitro hot spot-based antibody evolution. To evaluate non-germline hot spot residues as targets, we constructed two other libraries (RFB4LibVH30/31 and RFB4LibVH96/97/98) that are targeted to two non-germline hot spot regions in heavy chain CDR1 (Ser 30 -Ile 31 ) and CDR3

Anti-CD22 Immunotoxins
(Ser 96 -Gly 97 -Tyr 98 ), respectively. Ile 31 was mutated from Ser 31 in somatic hypermutation targeting the germline sequences around the RGYW (AGTAGCT) motifs (Fig. 1). Ser 96 -Gly 97 was generated by somatic insertion between V H and D gene fragments. The library complexity of RFB4LibVH30/31 consisted of more than 2 ϫ 10 4 independent clones. For library RFB4LibVH96/97/98, five independent transformations were performed to produce more than 1 ϫ 10 5 independent clones. Because theoretically the minimum library size for three hot spot residue randomization is 32,768 (32 ϫ 32 ϫ 32), the actual size of RFB4LibVH96/97/98 is 3-fold larger. Sequence analysis of 10 independent clones from each mutant library demonstrated that the targeted hot spot residues were successfully randomized (data not shown). As shown in Fig. 3, the pattern of enrichment of both libraries was strikingly different from the RFB4LibVL30/31 phage library. More rapid and early (at the second round) enrichment of binders with FACS signals similar to HA22 in library RFB4LibVH30/31 (Fig. 3B) indicated that it contained a large number of binders similar to the starting Fv HA22. But unlike RFB4LibVL30/31, the third round of RFB4LibVH30/31 panning failed to enrich the binders. In the case of RFB4LibVH96/97/98 (Fig. 3C), a large enrichment with FACS signals similar to HA22 occurred only at the third round of biopanning, suggesting that there are a very small number of binders in the phage-display library, and these binders are likely similar to HA22. The pattern of phage enrichment was also demonstrated in an ELISA on soluble CD22-Fc proteins (Fig. 4). Fig. 5 shows the CD22 binding activity of individual phage clones. Of 33 clones tested from the RFB4LibVH30/31 (2nd panel), 14 did not bind to CD22. Of the 19 binders, 9 had ELISA signals greater than the HA22 phage. Fig. 5, 3rd panel, shows the CD22 binding activity of phage clones from library RFB4VH96/97/98. Of 33 clones tested, only 7 bound CD22. Only 2 clones had ELISA signals slightly greater than the HA22 phage clone. ELISA-positive clones from RFB4LibVH30/ 31 and RFB4LibVH96/97/98 were sequenced (Table II). The sequences of the mutants enriched from the RFB4LibVH30/31 and RFB4LibVH96/97/98 were very similar to that of the parental antibody. Among the mutants of RFB4LibVH30/31, only one clone contains a mutation (Val 31 ) other than Ile/Leu 31 . It is also interesting to find that mutants selected from library RFB4LibVH96/97/98 targeting heavy chain CDR3 contained the similar sequence (Ser 96 -Gly 97 -Phe 98 ) (A3) or the same sequence (C4) as the parental one, Ser 96 -Gly 97 -Tyr 98 . Mutants H11 and E12 from LibRFB4VH30/31 were chosen to generate recombinant immunotoxins for further biological characterization. Cell death assays (Table IV) and protein synthesis inhibition assays (Fig. 7) showed that H11 and E12 were not more active than HA22. In fact, FACS analysis using the monomeric H11 and E12 immunotoxins showed that their binding to Daudi cells were similar to HA22 (Table IV). The results may suggest that the phage scFv binding assays are semi-quantitative, and a better indicator of successful enrichment may be that almost all the phage scFv clones bind the antigen strongly and that sequencing analysis shows that enrichment of new sequences.

DISCUSSION
The results presented here demonstrate that an evolution strategy using intrinsic mutational hot spots as targets can be used for antibody affinity maturation in vitro and therefore improve immunotoxin activity. More importantly, germline hot spots but not non-germline hot spots are good candidates for in vitro affinity maturation.
Germline Hot Spots and In Vitro Affinity Maturation-As shown in Table V, the number of germline RGYW hot spots tends to decrease in antibodies with high affinities, suggesting an overall correlation between affinity and number of hot spotbased somatic mutations that is characteristic of affinity maturation in vivo. The existence of germline hot spots in high affinity antibodies obtained in vivo may provide the possibility that in vitro affinity maturation targeting these hot spots could further increase the binding affinity. Therefore, we hypothesized that the germline hot spots could be good candidates for in vitro antibody affinity maturation. To test this hypothesis we first made a phage-display library targeting a germline hot spot located in light chain CDR1 encoding serine-asparagine. This library successfully produced several mutants in which Ser 30 -Asn 31 is most frequently replaced by either Arg 30 -Gly 31 or Gly 30 -Arg 31 . These Fv have affinities higher than the starting molecule, and immunotoxins made from them showed increased cytotoxic activity.
For comparison, we also made two phage-display libraries (RFB4LibVH30/31 and RFB4LibVH96/97/98) targeting nongermline hot spots in heavy chain CDR1 and CDR3. Two immunotoxins made from the H11 and E12 Fv of the RFB4LibVH30/31 library containing mutations in the nongermline hot spots did not show increased affinity and cytotoxic activity on Daudi cells (Table IV; Fig. 7). As shown in Table II, the sequences of the H11 and E12 mutants were very similar to TABLE IV Affinity and cytotoxic activity of RFB4 (scFv)-PE38 mutants Cytotoxicity was measured by cell death assays (WST-8). Briefly, cells were seeded at 5 ϫ 10 4 /well in a 96-well plate 24 h before the assay. Immunotoxins were added to the plate, and cells were incubated at 37°C for 20 h, and the cell viability was measured with WST-8. Each assay was done in triplicate. IC 50 (mean values expressed in ng/ml) is the toxin concentration that reduced cell viability by 50% compared with the cells that were not treated with the toxin. The results are represented as means Ϯ S.D. of triplicate determinations, and assays were repeated two or three times. In each CD22-positive cell line, repeated measures, analysis of variance with Dunnett's and Student-Newman-Keuls post-tests, revealed a significant increase in cytotoxicity (IC 50 ) of E6, B8, D8, and B5 as compared with BL22, HA22, and the H11 and E12 mutants (p Ͻ0.01 in Daudi and WSU-CLL cells, p Ͻ 0.05 in Namalwa and Ramos). The H11 and H12 mutants did not show increase of cytotoxicity as compared with HA22 (p Ͼ 0.05). No statistically significant differences among the E6, B8, D8, and B5 immunotoxins were observed (p Ͼ 0.05).  in antigen binding of RFB4. Together, these results may suggest that heavy chain CDR non-germline hot spot residues (Ile 31 , Ser 96 , and Gly 97 ), which have been mutated in vivo by a somatic process, are residues important for antigen binding and therefore may be unchanged by in vitro affinity maturation.
Furthermore, we analyzed mutants obtained from previous in vitro hot spot-based affinity maturation studies in our laboratory (21,27). In the case of first generation mutant HA22, germline hot spot residues Ser 100 -Ser 100A -Tyr1 00B in the D gene fragment (IGHD-FL16.1*01, GenBank TM accession number J00434) of heavy chain CDR3 were mutated to Thr 100 -His 100A -Trp 100B . SS is an anti-mesothelin single chain Fv antibody. SS1 is the mutant with an increased affinity selected by phage display (21). In SS1 germline hot spot Gly 93 -Tyr 94 in the light chain CDR3 (V germline sequences in GenBank TM accession numbers AJ235939, AJ231233, and AJ231211) is mutated to Lys 93 -His 94 . In contrast, when targeting a non-germline hot spot (Ser 53 -Thr 54 ) in the heavy chain CDR2 of the anti-Tac antibody in our laboratory, mutants enriched from the phage-displayed library have an affinity similar to the original Fv (data not shown). Almost all mutants contain Thr at the 54 position with nucleotide sequences different from the original clone, indicating true enrichment and the importance of Thr 54 in the antigen binding. It suggests that the phage-display library targeting a non-germline hot spot might simply regenerate the in vivo somatic hypermutation process in which the original germline hot spot Ser 53 -Ser 54 (IGHV1S27*01, Gen-Bank TM accession number X02459) was mutated into the nongermline hot spot Ser 53 -Thr 54 .
Most interestingly, in a recent study Wittrup and colleagues (19) created a random yeast surface-displayed scFv mutant library of humanized MFE-23 by mutagenic PCR. MFE-23 is an antibody specific for carcinoembryonic antigen. An scFv with a dissociation half-time over 4 days at 37°C was successfully isolated after four sorts. Two point mutations responsible for this affinity increase were identified and analyzed. The change that led to the greatest improvement in affinity is S50L of light chain. We compared the light chain of MFE-23 antibody with its germline counterpart (IgV ap4, GenBank TM accession number AJ231221). We found that Ser 50 is located in a germline hot spot sequence (AGC-ACA, Ser 50 -Thr 51 ) of the light chain CDR2 loop. The mutation of the germline hot spot residue (Ser 50 ) to Leu 50 appears to have a great effect on binding affinity and stability of the MFE-23 antibody. Based on the results described in this study and that of others, it is reasonable to expect that germline hot spots may prove valuable in antibody affinity maturation in vitro.
In Vitro Versus in Vivo Hot Spots-based Mutations, Different Mechanisms-We found that several mutated hot spot residues in the evolved antibodies arise from double or even triple mutations at the first, second, and third positions in each codon. This rarely occurs in the process of somatic hypermutation in B cells (23). For example, changing from AGC-AAG (Ser 30 -Asn 31 ) in the native mouse antibody to CAC-GGC (His 30 -Gly 31 ) in mutant B5 involves five nucleotide point mutations, in which all four nucleotides in the first and second positions are mutated. It is very improbable that such dramatic mutations can occur in vivo because it is well documented that hot spot mutagenesis in the somatic hypermutation process often involved only one nucleotide point mutation in each codon. In this case, Ser (AGC) can at most be mutated to only six residues in vivo: Cys (TGC), Gly (GGC), Ile (ATC), Thr (ACC), Asn (AAC), and Arg (CGC or AG(A/G)). Therefore, mutation from Ser 30 to His 30 as in the B5 mutant is rare. Similar dramatic change is also observed in mutant E6, in which Asn 31 is mutated into Arg 31 by changing AAG into CGG. In addition, HA22 has mutations of ACG-CAC-TGG (Thr-His-Trp) at the AGT-AGC-TAC (Ser-Ser-Tyr) hot spot motif in the CDR3 region. Most interestingly, the only double mutation at the first and second positions resulted in the presence of histidine in that mutant. This observation strongly suggests an advantage of in vitro hot spot-based antibody evolution as compared with in vivo somatic hypermutation. In contrast to the in vivo scenario, a hot spot residue can be easily mutated into other 19 amino acids in vitro by PCR. This may also explain why some hot spot mutations favorable for higher binding affinity fail to occur in vivo. In addition to constraints arising from the mechanism of somatic hypermutation, the B cells response exhibits an apparent affinity ceiling (Ͼ0.1 nM) during in vivo affinity maturation because of the inability to discriminate slower dissociation kinetics relative to intrinsic B cell receptor (BCR) internalization rates (38). Foote and Eisen (39) first argued that BCR (surface antibody)-antigen complexes with lifetimes much longer than the time necessary for internalization would all be processed equally well and hence could not distinguished. Indeed, Batista and Neuberger (38) demonstrated that relatively little benefit toward BCR internalization was achieved by increasing the half-life of antigen/BCR TABLE V Number of germline RGYW hot spots and affinity maturation Top, immunotoxins containing the Fv portions of RFB4, SS, and Anti-Tac are currently under investigation in our laboratory for affinity maturation. The original Fv (boldface) and its mutants obtained by in vitro hot spot mutagenesis are shown. Affinities (K D values) were measured by flow cytometry (RFB4 and its mutants) or BIAcore (SS, SS1, and Anti-Tac) by using recombinant immunotoxins. Bottom, three independent monoclonal antibodies (H26, H10, and H8) from mice immunized with the same protein antigen (hen egg white lysozyme) that recognize overlapping epitopes on the antigen (42,43). All three antibodies use the same V L (Igk-v23) and V H (VHM460) germline genes. K D values were determined by BIAcore using Fab as described (44 Anti-CD22 Immunotoxins dissociation beyond 30 min. In the present study, in the final round of biopanning on Daudi cells expressing the CD22 antigen in the presence of 10-fold excess of phage antibodies, the total duration of the off-rate selection was 6 h, a far longer interval than the 30-min window. In order to inhibit the internalization of CD22, we used NaN 3 and low temperature during the cell panning. However, we were not able to increase the affinity to 0.1 nM or better. It is possible that we would have selected a better mutant if we used more (100-fold) competitors in the panning (18), longer off-rate selection, or more rounds of enrichment.
What is more likely is that a mutant with a K D of 0.1 nM or less simply does not exist in the current hot spot library and therefore further stepwise evolutionary process targeting hot spots in other CDR regions of RFB4 (e.g. L-CDR3) may be needed. Linkage of Affinity with Cytotoxicity of Immunotoxins-In previous works, we found a discrepancy between an increase in affinity and cytotoxicity when panning or measuring the affinities by using recombinant protein made in E. coli or peptides (21,26). In the present study, we initially performed phage panning on recombinant CD22-Fc proteins with the phagedisplay library RFB4LibVL30/31. Mutants containing Ser 30 -Lys 31 and Ser 30 -Gly 31 in light chain CDR1 were successfully enriched after the third round panning. Several clones with the same amino acid (Ser 30 -Lys 31 ) but different nucleotide sequences were found. A Ser 30 -Lys 31 mutant (A1E) was chosen for further characterization. It bound to the CD22-Fc proteins with an ELISA signal 3-fold higher than HA22. But when tested on CD22-positive cells, a recombinant immunotoxin containing the A1E Fv had cytotoxic activity only slightly (ϳ30%) better than HA22 on Raji cells and almost the same as HA22 on Daudi cells (data not shown). Because of that, we decided to use cells expressing native CD22 molecules on the cell surface for both biopanning and affinity determination. It appeared to improve the linkage of phage-displayed antibody binding affinity with immunotoxin cytotoxicity. By doing that, we should not have missed mutant antibodies recognizing conformational or carbohydrate epitopes. It may be especially important in this study because the precise epitope of the RFB4 antibody on CD22 has not been elucidated.
Structural Implications of Mutations-The formation of antibody-protein antigen complexes involves a combination of electrostatic and hydrophobic interactions at the interface between the two proteins (40). First generation mutant HA22 replaces the Ser 100 -Ser 100a -Tyr 100b (Fig. 2B) with Thr 100 -His 100a -Trp 100b on the heavy chain CDR3 loop. Second generation mutants enriched from phage-display library RFB4LibVL30/31 replace Ser 30 -Asn 31 with Gly 30 -Arg 31 (E6) and Arg 30 -Gly 31 (B8) in light chain CDR1 (Table II). In both cases, positively charged residues were selected. In fact, we previously found that a single histidine can increase the affinity by 3-fold (27). It is reasonable to expect that the surfaceexposed positions of basic residues may strengthen the electrostatic interactions at the interface between the RFB4 antibody with the CD22 antigen.
In summary, the uphill-climbing approach using intrinsic germline hot spots successfully evolved several immunotoxins with increased activity on B lymphoma cells including CLL, strongly suggesting that they should be very useful in the treatment of CLL and other cancer that express low levels of CD22. Furthermore, potential applications of this germline hot spot strategy extend beyond antibody affinity maturation and improvement of immunotoxin activity to the modification of receptors, ligands, and other biologically important proteins because somatic hypermutation may be a global phenomenon in the whole genome, and similar DNA hot spot mechanisms are not restricted to immunoglobulin genes (41).